In many of the papers it is said that the derivative filter transfer function is given by: $$H(z) = \dfrac{1}{8T}\left(-z^{-2} - 2z^{-1} + 2z + z^{2}\right)$$ But no one gave the detailed information about it. Anyone has an idea about this?

  • $\begingroup$ Please cite the specific paper so we can look at the contest. It's one of the many possible ways to build an approximate differentiator $\endgroup$
    – Hilmar
    Jul 1 '15 at 19:15
  • $\begingroup$ academia.edu/5263479/…. In this paper they used 5 point derivative filter. But I am still banging my head with transfer function $\endgroup$ Jul 1 '15 at 19:44

This is not a standard 5-point derivative formula, the corresponding transfer function of which is


The figure below shows the magnitude responses of an ideal differentiator (red), of the standard 5 point approximation given by (1) (green), and of the approximation in your question (blue):

enter image description here

As you can see, the non-standard approximation approximates the ideal differentiator only in a frequency region up to about $1/10$ of the Nyquist frequency (half the sampling frequency), whereas the standard approximation approximates the ideal differentiator quite well up to about a normalized frequency of $0.3$.

The differentiator in your question is a maximally flat lowpass differentiator. It avoids amplification of high-frequency noise. It achieves a maximally flat approximation of the ideal differentiator at $\omega=0$ and it has a given number of zero derivatives at $\omega=\pi$ (Nyquist), i.e. it is has a flat response at Nyquist, which causes high frequencies to be attenuated.

The design formula is derived in this paper. There is also a Matlab program available to design such maximally flat differentiators: link. Your specific filter is obtained by the command h=lowdiff(3,0);

  • $\begingroup$ Nice. $ a \Delta _{\pm 1} \ + (1-a) \Delta _{\pm 2} ), \ a \ 1/2 .. 4/3, $ traces out curves in between (with noise amplification factors 0.4 .. 0.95). $\endgroup$
    – denis
    Jan 18 '19 at 11:44

It seems to be some kind of weighted central-difference formula over different time intervals, maybe to minimize the effect of noise on the signal for shorter time intervals.

In your case the weighted central-difference formula can be written as,

$$ \dot{y}(t) \approx \frac{\frac{y(t+2T) - y(t-2T)}{4T} + \frac{y(t+T) - y(t-T)}{2T}}{2} = \frac{-y(t-2T) - 2y(t-T) + 2 y(t+T) + y(t+2T)}{8T}. $$


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.